A Tool for Predicting the Success of First-Line Antiretroviral Therapies Alejandro Pironti Computational Biology and Applied Algorithmics Max Planck Institute for Informatics April 28, 2010
Motivation • The first antiretroviral regimen: – Best chance of sustained virological suppression – Selection under consideration of transmitted drug resistance – Simplicity and side effects also important for success Primary Drug Resistance Trends in the RESINA Study 20 Prevalence(% ) 15 10 5 0 2001 2002 2003 2004 2005 2006 2007 2008 Year Alejandro Pironti April 28, 2010
Motivation • Available tools for assisting therapy selection: – Designed for therapy-experienced patients – Therapy-naïve patients: • Different mutations, e.g. T215Y → T215C • A subset of regimens Alejandro Pironti April 28, 2010
The Data • 2401 first-line antiretroviral therapies from EuResist and the RESINA study including: Counts Alejandro Pironti April 28, 2010
The Data • 2401 first-line antiretroviral therapies from EuResist and the RESINA Study including: – PR and RT sequences – Various viral load measurements • Linear interpolation of log VLs to obtain values for week 36 Viral Load Measurements by Week Alejandro Pironti April 28, 2010
Dichotomization • Failure: 70 – VL > 100 cp/ml at week 36 60 – Therapy stop before week 50 36 • Definition yielded 40 % – 1527 (64%) successes 30 – 874 (36%) failures 20 • Development set: 1981 10 therapies 0 • Test set: 483 therapies Virological Failure at Week 36 Other Failure Success Alejandro Pironti April 28, 2010
Prediction Schematic Representation of • Linear support vector a Support Vector Machine machine was used to predict therapy success n i g r a M • Training with development set • Features: – Genotype – drug combination – drug interactions up to 3 rd order Hyperplane • Feature selection: z-score ≥ 2 Alejandro Pironti April 28, 2010
Performance 5-fold cross-validation on Test set. AUC = 0.7110 development set. AUC = 0.7135 EuResist Engine acheives AUC=0.5620 on the test set Alejandro Pironti April 28, 2010
Selected Features • 70 out of 1280 features selected • Protease Mutations: – T12K I15L L19P K20V N37E C67S E79D T91A T91V I93M • Reverse Transcriptase Mutations: – K43R K43N A62V D67N L74V V90I A98G K102E Y144F K173V Q174N N175Y V179D D192N P197E T200E E203D Q207K R211A R211E T215C T215E D218E K219Q W229G K238N • 7 single drugs, 23 two-way drug interactions and 4 three-way drug interactions Alejandro Pironti April 28, 2010
Tool Development Status • Core algorithm of the tool has been implemented • Web interface will be implemented and made available online by July 2010 • Tool will be part of geno2pheno[resistance] Alejandro Pironti April 28, 2010
Example • Case from the test set • AZT 3TC LPV/r • RT substitutions: – 68G, 121H, 135T, 158S, 173Q, 200K, 211K, 214F, 215C • PR substitutions: – 3I, 12N, 12K, 19I, 35D, 37N, 63P, 77I • Viral load > 100 cp/ml at week 36 Alejandro Pironti April 28, 2010
Example Geno2pheno[resistance] output • Enter sequence into geno2pheno[resistance] • RT substitutions: – 68G, 121H, 135T, 158S, 173Q, 200K, 211K, 214F, 215C • PR substitutions: – 3I, 12N, 12K, 19I, 35D, 37N, 63P, 77I • WHO transmitted drug resistance surveillance mutations: 215C Alejandro Pironti April 28, 2010
Example Compounds Success Probability TDF FTC SQV/r 0.82 TDF FTC EFV 0.76 TDF FTC ATV/r 0.76 LPV/r ABC 3TC 0.76 TDF FTC NVP 0.76 ABC 3TC SQV/r 0.76 TDF FTC LPV/r 0.76 ABC 3TC EFV 0.76 ABC 3TC NVP 0.75 AZT 3TC LPV/r 0.74 AZT 3TC FPV/r 0.73 Alejandro Pironti April 28, 2010
Acknowledgements University of Cologne Max-Planck-Institut für Informatik Rolf Kaiser Thomas Lengauer Marc Oette André Altmann Melanie Balduin Joachim Büch Saleta Sierra Aragon Finja Schweizer Elena Knops Maria Neumann-Fraune University of Erlangen Eugen Schülter Hauke Walter Eva Heger University of Düsseldorf Björn Jensen Alejandro Pironti April 28, 2010
Recommend
More recommend
